StephenShumway/rigless-lora

Rigless LoRA v1

A LoRA for MiniMax H3 that replaces a person in real footage with a character from pictures. It was made for Rigless, which downloads it with the other models. The file is rigless-lora-v1.safetensors.

What it is

This is a modified version of akatz's H3 Relay Character Swap LoRA v1.5: v1.5 fine-tuned further on real footage. Each training sample is a real clip with someone else swapped in by H3. The LoRA learns to turn that back into the real person, given only a picture of them. The target is always real footage, so it learns real light and has to read the picture to know who to draw.

In a blind test it was preferred over v1.5 and over no swap LoRA (3 clips, one take each, judged by its developer). It is not measurably better at face identity.

How to use it

With Rigless, nothing to do: it downloads this LoRA and loads it for every render.

In ComfyUI, put it in models/loras and load it at strength 1.0 on MiniMax H3 (ref2va), with the H3 Relay nodes and the 8-step H3 Turbo LoRA, as Rigless's graph in src/rigless/swap.py does. It was trained with this prompt for one picture:

Replace the selected person in <Video 1> with the person in <Picture 1>. Keep the reference person's face, hair and clothing. Follow the source performance's pose, gaze, expression and movement. Preserve the scene and the other people.

License

The MiniMax H3 Community License Agreement, in LICENSE, with its NOTICE. This LoRA is a derivative of the MiniMax H3 Works and comes with no terms beyond that license. MiniMax H3 can't be used in the United States, the European Union, the United Kingdom or South Korea without your own authorization from MiniMax. Use footage and pictures you have the rights to, of people who consented or who aren't real, and say clearly that a video is AI-generated wherever you share it.

Credits

MiniMax for MiniMax H3, and akatz for H3 Relay and the Character Swap LoRA v1.5 this LoRA is built on.